Make Dataset the first class interface

[GeorgesLorre]

Given our vision we want to promote the idea of a dataset.

Now fondant is very pipeline focused where we see a pipeline as a graph of operations with intermediate datasets. We should rethink the our primary interface to enable the dataset first approach:

We should pack more functionality in the Dataset class:

• view data preview (html formatted like pandas)
• view data schema / metadata
• view lineage
• ...

We should abstract the Pipeline idea. Compiling and starting a pipeline is a cheap operation so it should be less static. If we store the correct information on the dataset class we can create pipelines from a dataset (and its dependencies).

Tasks

Beta Give feedback

1. Initiliase dataset from previous Fondant run #885
   +0
   
2. Refactor Pipeline interface #886
   Core +1
   
3. Add workspace context #887
   +0
4. Remove pipeline references in core code #900
   +0
5. Update documentation #909
   +0
6. Update Manifest #910
   +0
7. Update kubeflow and vertex compiler #913
   +0